[Paper Review] A computational model of affects

Mika Turkia|ArXiv.org|Nov 2, 2008

Evolutionary Algorithms and Applications9 references3 citations

TL;DR

This paper proposes a computationally defined model of affects as unconscious physiological processes triggered by specific events, differentiating them by event type (expected/unexpected, past/future, self/other-related). It restructures the OCC model into a fully implementable framework using utility-based event evaluation and object modeling, enabling simulation of complex affective dynamics like pride, guilt, envy, and mood states in agents with minimal cognitive overhead.

ABSTRACT

This article provides a simple logical structure, in which affective concepts (i.e. concepts related to emotions and feelings) can be defined. The set of affects defined is similar to the set of emotions covered in the OCC model (Ortony A., Collins A., and Clore G. L.: The Cognitive Structure of Emotions. Cambridge University Press, 1988), but the model presented in this article is fully computationally defined.

Motivation & Objective

• To address the lack of computational clarity in existing emotion models like OCC, which rely on undefined concepts such as norms and standards.
• To define affects as unconscious, evolutionarily grounded physiological state changes triggered by specific events, enabling direct implementation in artificial agents.
• To provide a logical, minimal framework for affective states that supports both innate and learned behavioral responses.
• To demonstrate how social and self-referential affects—such as pride, guilt, envy, and mood—emerge from utility-based event evaluation and object modeling.
• To enable applications in computational psychology, sociology, psychiatry, and interactive systems like computer games.

Proposed method

• Affects are defined as unconscious, evolutionarily determined changes in body state triggered by perception of events that alter utility relative to an agent’s goals.
• Agents maintain an object model of past events, associating them with utility values to predict future outcomes and form expectations.
• Affect types are differentiated by event type: unexpected positive/negative/neutral events trigger delight/surprise/fright; expected future events trigger hope/fear; fulfilled/unfulfilled expectations trigger satisfaction/disappointment/fear-confirmed/relief.
• Self-originated events trigger pride (positive) or shame (negative); actions affecting liked/disliked agents trigger guilt (negative action on liked), pity (negative on liked), resentment (positive on disliked), or gloating (negative on disliked).
• Agent valence (like/dislike) is determined by net utility of past interactions; desired/disliked status is based on expected future utility.
• Mood is modeled as a function of average utility across events, with happiness (positive average), sadness (negative average), and depression (no positive utilities).

Experimental results

Research questions

• RQ1 How can affective states be formally and computationally defined without relying on undefined psychological concepts like norms or standards?
• RQ2 In what way can unconscious affects alone generate complex behavioral and emotional responses in artificial agents?
• RQ3 How do expectations about future or past events give rise to specific affective states such as hope, fear, satisfaction, or relief?
• RQ4 How can social affects like envy, gratitude, or resentment emerge from utility-based interactions between agents?
• RQ5 How can mood be modeled as a stable, long-term affective state derived from cumulative event history?

Key findings

• The model successfully defines a comprehensive set of affects—including pride, guilt, envy, and mood—using only utility changes and event expectations, without relying on undefined concepts.
• The system demonstrates that complex social affective dynamics, such as pity, gloating, and gratitude, emerge naturally from simple utility-based interactions and object modeling.
• In the browser-based simulation, agents developed stable expectations and exhibited affective responses consistent with psychological theory, such as disappointment after unmet expectations and relief after avoided negative outcomes.
• The model supports both innate and learned behavioral associations, such as aggression in frustration or identification with high-utility agents.
• Agents exhibited mood shifts in response to cumulative utility patterns, with depression emerging when no positive utilities were available and bad mood when expectations turned negative.
• The simulation confirmed that self-referential affects like pride and remorse arise from self-generated actions and their outcomes, with remorse triggered by negative consequences of self-initiated actions.